g95 runtime errors

[Matt]

Hi guys.  It's been a while since I needed help from the group (my
CP2K research has been put to the side since I came to Africa and lost
my access to large computing clusters).  I'm trying to build and run
CP2K on some older machines that were donated, in hopes that I could
use it to teach my Master's students about DFT and MD.

I am compiling on Fedora Core 5 with g95 and using the Linux-i686-g95
arch file (from what I can tell, x86, i686, and ia-32 are all the same
architecture, correct?  I'm running on Celerons and Pentium 4s).  I
use the arch file directly from the CVS repository, but change the
blas and lapack library location.  The code compiles with no problems
(I've also compiled using FFTW3 with no problems).

The problems arise when I try to do the regtesting.  Every test fails
with a segmentation fault.  I've been table to trace it to the initial
reading of the input file.  In create_cp2k_input_reading/input_cp2k.F,
the res%section pointer is not associated after calling
section_vals_create, even though input_section is associated.  It
seems to me that res%section should be pointed to input_section during
section_vals_create.  Indeed, it is, but for some reason the program
is not doing it.  The recursiveness of these subroutines has prevented
me from finding out exactly why it's not getting associated, and when
this happens.  Because it's not associated, the program segmentation
faults in the section_vals_parse call a few lines later.

section%repeats is not associated (around line 146)
section_vals_parse in input_parsing.F
called by create_cp2k_input_reading in input_cp2k.F
called by cp2k_run in cp2k_runs.F

Looking through the Group, this problem hasn't been encountered before
(seems like g95 is usually stable after it's compiled).  Has anyone
seen anything like it or can give me a clue to what's going on?

Please excuse me if I don't respond quickly.  The Internet/electricity
isn't always working here, so I may not be able to check my email.

                                        Cheers, Matt